Electrocardiogram classification - A human expert way

Research output: Contribution to journalConference articleScientificpeer-review

Researchers

Research units

  • RemoteA Ltd

Abstract

We present an easy-to-understand classifier for the PhysioNet/Computing in Cardiology Challenge 2017. The classifier mimics the workflow of a human expert in classifying atrial fibrillation and other cardiac arrhythmias based on short single lead electrocardiogram. No computational methods were used for defining or tuning the classification rules. The ECG data was preprocessed by running a custom made beat detection and clustering algorithm. Samples of preprocessed data were then shown to a human expert, who was asked to define rules for classifying the data into subsets. The resulting one-sided binary tree classifier scored 73 % in a hidden subset. Our goal was to study how well simple human understandable rules are able to compete against advanced classification system - they are compatible, but at least our approach was clearly behind the top score in the competition (83 %).

Details

Original languageEnglish
Pages (from-to)1-4
Number of pages4
JournalComputing in Cardiology
Volume44
Publication statusPublished - 2017
MoE publication typeA4 Article in a conference publication
EventComputing in Cardiology Conference - Rennes, France
Duration: 24 Sep 201727 Sep 2017
Conference number: 44

ID: 30474805